Adding a python_requires < 3.7

[georgealton]

Would adding a python_requires = '<3.7', configuration to the setup.py, to prevent this package from being installed in python environments >=3.7, be helpful?

This would help to prevent accidental shadowing of the builtin dataclass package, and then triggering some of the incompatability issues

[ericvsmith]

Does that actually work? It's an honest question: I never specify versions in setup.py, only in requirements.txt.

If I'm installing a 3.7 venv with a requirements.txt that specifies dataclasses, with no python_requires, will pip not install dataclasses if its setup.py says python_requires = '<3.7'?

[georgealton]

That's my interpretation of the option from the guide; https://packaging.python.org/guides/distributing-packages-using-setuptools/#python-requires

I just had a go at creating that scenario - pip refused to install my local modified version of dataclasses with the marker;

Here usage of this functionality in hypothesis https://github.com/HypothesisWorks/hypothesis/blob/f6642662f9f56687817e49c6fd25c30e79d2b2a6/setup.py#L77

[edo248]

It indeed works, by adding data value in the pip index which pip honors. You can see this data by looking at the source of index for some package. Even for pip itself, check source code of https://pypi.org/simple/pip

An example lines looks like this:

<a href="https://files.pythonhosted.org/....." 
      data-requires-python="&gt;=2.7,!=3.0.*,!=3.1.*,!=3.2.*,!=3.3.*,!=3.4.*">
                 pip-19.2.tar.gz
</a>
<br/>

[ericvsmith]

I just released 0.7 on pypi to address this issue. Sorry for the long delay.

[luckydonald]

With typing already being present in python 3.5, why is this backport 3.6 only?

[ericvsmith]

Because variable annotations (PEP 526) didn't show up until 3.6, and dataclasses requires them.

[luckydonald]

Oh I thought it would only be the format strings

f"Some {variable!r}"

which could be "downgraded" to

"Some {variable!r}".format(variable=variable)

Thanks for clarifying.